The productivity of milling machines is limited by chatter vibrations. Stability lobe diagrams (SLD) allow the selection of suitable process parameters to maximize the productivity. However, the calculation of SLDs is very time-consuming and requires complex experiments. In this article a new online learning method is presented, which allows the calculation of SLDs during the production process. The algorithm is a combination of reinforcement learning and nearest-neighbor-classification and allows the estimation of the stability border based on measured vibration signals during machining. The proposed algorithm is capable of being continuously trained with sorted input data. A trust criterion is introduced, which allows judging the prediction quality of the algorithm. The algorithm is validated with analytical benchmark functions and with a 2-DOF milling stability simulation.